Meta’s codenamed Watermelon model, the successor to Muse Spark, remains in advanced training or post-training as of early October 2026 after missing an anticipated Meta Connect reveal in September. Internal statements from July indicated the frontier-scale system had reached parity with OpenAI’s GPT-5.5 on key benchmarks while consuming roughly ten times the compute of its predecessor, reflecting Meta’s aggressive infrastructure push. Reports from August cited an internal October target for public release, positioning the model for text-based arena evaluations that would allow direct capability comparisons. Recent Muse Spark updates in September have delivered incremental coding and agentic gains, but traders are watching for any verified Watermelon weights, API access, or independent arena results that could confirm or adjust competitive positioning against OpenAI, Anthropic, and Google models.